Grid Federation: Number of Jobs and File Size Effects on Jobs Time

Abstract

Grid federation is fast emerging as an alternative solution to the problems posed by the large data handling and computational needs of the existing numerous worldwide scientific projects. Efficient access to such extensively distributed data sets has become a fundamental challenge in grid computing. Creating and placing replicas to suitable sites, using data replication mechanisms can increase the system’s performance. Data Replication reduces data access time, ensures load balancing as well as narrows bandwidth consumption. In this paper, an enhanced data replication mechanism called EDR is proposed. EDR applies the principle of exponential growth/decay to both file size and file access history, based on the Latest Access Largest Weight (LALW) mechanism. The mechanism selects a popular file and determines an appropriate number of replicas as well as suitable grid sites for replication. It establishes the popularity of each file by associating a different weight to each historical data access record. Typically, recent data access record has a larger weight, which signifies that the record is more relevant to the current situation of data access. By varying the number of jobs as well as file sizes, the proposed EDR mechanism was simulated using file size and job completion time as the variable metrics. Optorsim simulator was used to evaluate the proposed mechanism alongside the existing Least Recently Used (LRU), and Least Frequently Used (LFU) Mechanisms. The simulation results showed that job completion time increases by the growth in both file size and number of jobs. EDR shows improved performance on the mean job completion time, compared to LRU and LFU mechanisms.